ARTIFICIAL INTELLIGENCE DETECTS BREAST CANCER UP TO TEN YEARS BEFORE DIAGNOSIS - WHAT A STUDY BY GREEK SCIENTISTS SHOWED - Filenews 23/8
AI systems can detect early signs of breast cancer up to six years before diagnosis and, in some cases, even a decade earlier, according to a study involving Greek scientists.
The research, which was published in the scientific journal "Radiology" of the Radiological Society of North America, paves the way for the earliest detection of the disease.
In the retrospective study, the researchers evaluated three commercially available and reliable AI systems, which they fed with data from mammograms in Sweden. Specifically, the study included almost 89,000 mammograms from 31,394 women, performed over a period of about ten years, between January 2008 and April 2019. During this period, 12,072 women, or 38.5% of the sample, were diagnosed with cancer by radiologists.
The data came from the "Validation of Artificial Intelligence for Breast Imaging" database, which collects breast imaging data from volunteers in four regions of Sweden. It is noted that in the country's national screening program, women aged 40 to 74 are asked to undergo a mammogram every two years and each mammogram is traditionally evaluated by two radiologists.
The researchers found that about 20% of breast cancer cases showed signs on mammograms that were already visible by AI systems much earlier than radiologists detected them. In 20% of women, the systems recognized signs six years before diagnosis, in 25% four years earlier, and in 39% two years before diagnosis. In about 15% of women, the systems detected cancers even ten years earlier, although in this time range there were small differences between the three systems. In addition, the systems achieved a very high ability to distinguish between truly positive and negative results of 90%.
The research was carried out by the radiologist, Pantelis Gialias, as part of his doctoral studies at the Swedish University Hospital Linkoping. The Greek researcher from the Swedish Karolinska Institute, Apostolia Tsirikoglou, also participated in the research.
As Mr. Gialias explains, the cancer cases detected by artificial intelligence systems "are underlying changes in which the human eye cannot say with certainty that there is something suspicious. It can be a small disorder of breast architecture or a gradual increase in breast density, which are small signs that you can't rely on no matter how experienced a radiologist you are."
Originally from Chios and studying at the Medical School of the Comenius University of Bratislava and specializing in Radiology in Greece, Mr. Gialias specialized from a very early age in breast imaging and obtained a relevant certification in Sweden. It describes how dramatically the effectiveness of Artificial Intelligence in Radiology has changed in a decade. "When we first tested the usefulness of Artificial Intelligence in detecting breast cancer in 2015, the results were disappointing. However, over the years the algorithms have improved and computers have become much more powerful to process a huge volume of tests, so the help they can offer has been shown."
In a previous research conducted by Mr. Gialias in 2022 and published in "Acta Radiologica", it was identified that Artificial Intelligence can reduce the workload of radiologists in the study of mammograms by up to 34%. In a subsequent research published last year in "European Radiology", it was found that the use of digital mammography with artificial intelligence during screening is also a cost-saving strategy compared to the traditional method of examining the results by two radiologists.
"When Artificial Intelligence first appeared, there was a fear that radiologists would be left without a job. This is not the case. Essentially, it is an additional tool, a weapon we have for the early diagnosis of breast cancer. Of course, all decisions must be made with care and with the necessary studies and there must always be human control," emphasizes Mr. Gialias, who is currently the director of the Breast Center of the Cypriot Mediterranean Hospital.
He explains that the breast is a particularly complex organ, a gland, whose image changes significantly even in the same woman, depending on hormonal changes. The bet now, she adds, is "to see whether we can use artificial intelligence systems under certain protocols so that women who show the early signs follow a different pattern of screening on a case-by-case basis, with the aim of detecting cancers much earlier."
At the same time, he emphasizes the need to conduct prospective studies, while the research team plans to continue the study focusing on women with silicone implants in the breasts, where imaging has additional challenges.
RES-EAP / Radiological Society of North America (RSNA)
AI systems can detect early signs of breast cancer up to six years before diagnosis and, in some cases, even a decade earlier, according to a study involving Greek scientists.
The research, which was published in the scientific journal "Radiology" of the Radiological Society of North America, paves the way for the earliest detection of the disease.
In the retrospective study, the researchers evaluated three commercially available and reliable AI systems, which they fed with data from mammograms in Sweden. Specifically, the study included almost 89,000 mammograms from 31,394 women, performed over a period of about ten years, between January 2008 and April 2019. During this period, 12,072 women, or 38.5% of the sample, were diagnosed with cancer by radiologists.
The data came from the "Validation of Artificial Intelligence for Breast Imaging" database, which collects breast imaging data from volunteers in four regions of Sweden. It is noted that in the country's national screening program, women aged 40 to 74 are asked to undergo a mammogram every two years and each mammogram is traditionally evaluated by two radiologists.
The researchers found that about 20% of breast cancer cases showed signs on mammograms that were already visible by AI systems much earlier than radiologists detected them. In 20% of women, the systems recognized signs six years before diagnosis, in 25% four years earlier, and in 39% two years before diagnosis. In about 15% of women, the systems detected cancers even ten years earlier, although in this time range there were small differences between the three systems. In addition, the systems achieved a very high ability to distinguish between truly positive and negative results of 90%.
The research was carried out by the radiologist, Pantelis Gialias, as part of his doctoral studies at the Swedish University Hospital Linkoping. The Greek researcher from the Swedish Karolinska Institute, Apostolia Tsirikoglou, also participated in the research.
As Mr. Gialias explains, the cancer cases detected by artificial intelligence systems "are underlying changes in which the human eye cannot say with certainty that there is something suspicious. It can be a small disorder of breast architecture or a gradual increase in breast density, which are small signs that you can't rely on no matter how experienced a radiologist you are."
Originally from Chios and studying at the Medical School of the Comenius University of Bratislava and specializing in Radiology in Greece, Mr. Gialias specialized from a very early age in breast imaging and obtained a relevant certification in Sweden. It describes how dramatically the effectiveness of Artificial Intelligence in Radiology has changed in a decade. "When we first tested the usefulness of Artificial Intelligence in detecting breast cancer in 2015, the results were disappointing. However, over the years the algorithms have improved and computers have become much more powerful to process a huge volume of tests, so the help they can offer has been shown."
In a previous research conducted by Mr. Gialias in 2022 and published in "Acta Radiologica", it was identified that Artificial Intelligence can reduce the workload of radiologists in the study of mammograms by up to 34%. In a subsequent research published last year in "European Radiology", it was found that the use of digital mammography with artificial intelligence during screening is also a cost-saving strategy compared to the traditional method of examining the results by two radiologists.
"When Artificial Intelligence first appeared, there was a fear that radiologists would be left without a job. This is not the case. Essentially, it is an additional tool, a weapon we have for the early diagnosis of breast cancer. Of course, all decisions must be made with care and with the necessary studies and there must always be human control," emphasizes Mr. Gialias, who is currently the director of the Breast Center of the Cypriot Mediterranean Hospital.
He explains that the breast is a particularly complex organ, a gland, whose image changes significantly even in the same woman, depending on hormonal changes. The bet now, she adds, is "to see whether we can use artificial intelligence systems under certain protocols so that women who show the early signs follow a different pattern of screening on a case-by-case basis, with the aim of detecting cancers much earlier."
At the same time, he emphasizes the need to conduct prospective studies, while the research team plans to continue the study focusing on women with silicone implants in the breasts, where imaging has additional challenges.
RES-EAP / Radiological Society of North America (RSNA)
